Why should you care?

You need evidence from production with low overhead when something intermittent breaks.

See it live — copy this example

Run in SQL Server Management Studio (SSMS) or Azure Data Studio.

-- Create a simple error ring buffer (run once)
CREATE EVENT SESSION [DataVerse_Errors] ON SERVER
ADD EVENT sqlserver.error_reported (
    ACTION(sqlserver.sql_text, sqlserver.client_hostname)
    WHERE [severity] >= 16
)
ADD TARGET package0.ring_buffer
WITH (STARTUP_STATE = OFF);
ALTER EVENT SESSION [DataVerse_Errors] ON SERVER STATE = START;
-- Later: ALTER EVENT SESSION [DataVerse_Errors] ON SERVER STATE = STOP;

What happened?

• The session catches severe errors with SQL text into a ring buffer.
• Start it when investigating; stop it when done to avoid noise.

Practice next

1. Create and start your database_Errors in lab.
2. Cause a handled error (divide by zero) and inspect targets.
3. Add xml_deadlock_report for deadlock graphs.
4. Add rpc_completed WHERE duration > 1000000 (microseconds).
5. Export deadlock XML and open in SSMS.

Remember

XEvents = modern tracing. Filter aggressively. Start/stop with intent.

Deadlock graph capture

DataVerse prod runs a deadlock XEvent session.

Outcome: Engineers fix lock order using real graphs, not guesses.